In Regulated Work, the Agent Product Is Becoming the Guardrail

Healthcare calls arrive faster than policy

Pindrop reports that weekly calls from Meta Muse AI agent grew eightfold in its data between September 9 and October 6, 2026, and that healthcare received the second-highest volume of AI agent calls among analyzed industries, behind financial institutions. Providers use those agents to check benefit eligibility, claim status and prior authorizations, work that an American Medical Association survey puts at an average of 40 prior authorizations per physician per week, consuming about 13 hours of physician and staff time. m:…7389

The same traffic creates the risk: Pindrop says malicious bots conduct reconnaissance to map interactive voice response systems, discover authentication gates and collect patient information, and that bad actors use bots to compile sensitive data for larger attacks leading to privacy breaches and financial loss. Pindrop recommends validating authenticity, identity and intent for every agent caller, notes that 94 percent of analyzed agent calls were labeled good, and reports that JPMorgan Chase and Charles Schwab have implemented policies to block unauthorized agents, while Guardian Life keeps a human in the loop while moving toward governance that flags exceptions. m:…7389

Compliance moves into the agent runtime

Sierra built Compliance Console, integrated into its Horizon platform, to provide guardrails for agent compliance as agents shift from reactive support to long-horizon interactions lasting weeks or months. Regulatory rules for customer interactions vary by country, state and industry, and the company warns non-compliance can bring fines, legal action and loss of customer trust, which is why the console lets teams configure disclosures for calls, texts and emails, set contact windows around the recipient local time, and hold outreach automatically on specific regional holidays. m:…0852

The console also supports separate frequency limits for marketing and transactional outreach, saves opt-out requests so they remain accessible to the company, and shows dashboards counting calls and messages allowed, deferred, blocked or held for review. The standard acceptable contact window is 8AM to 9PM, a small detail that illustrates the larger point: compliance for agents is a configuration surface with defaults, overrides and an audit trail, not a policy document. m:…0852

Legal and clinical intake follow the same shape

Harvey launched an Intake artifact for Claude that lets business teams ask questions and submit requests inside Claude, connected to Harvey through a server that mediates the link, with every request logged for tracking and triage. Routine legal questions are answered using the legal team guidance, playbooks and prior work stored in Harvey, while requests requiring legal judgment are routed to a lawyer, and requesters can check status and follow up without leaving Claude. m:…2369

OpenEvidence and the American Society of Addiction Medicine announced a partnership on October 28, 2026, making ASAM clinical guidelines and resources accessible through the OpenEvidence platform, where the guidance surfaces automatically when clinically relevant. The context is stark: 44.6 million Americans had a substance use disorder in 2025, according to the announcement, and ASAM represents more than 8,000 physicians and associated professionals. In both cases the agent earns trust by staying inside a governed source, not by answering more freely. m:…2447

The buyer checklist

A buyer evaluating any of these products can now ask a short list of questions drawn directly from the releases: can the system validate the authenticity, identity and intent of every agent caller, as Pindrop recommends, and can it show a dashboard of outreach allowed, deferred, blocked or held for review, as Sierra provides. If a vendor cannot answer both questions with a log and a dashboard, the agent may work well in a demo while leaving the regulated buyer to carry the compliance risk alone. m:…7389 m:…0852